Study On The Analysis Methods Of Vibration Mode For A Fault Rotor Based On HHT

This thesis, in which the research work focused on the topic of faulty vibration mode analysis of rotor system, was mainly supported by the High Technology Research and Development Program of China’Research on the condition monitoring and fault diagnosis technique and its application software system of supercritical and ultra-supercritical steam turbine generator unit’(No.2008AA04Z410) and the National Natural Fund Project’Research on the analysis of vibration mode for rotating machinery under typical fault based on EMD’(No.11072214). This study consists of the definition of the vibration mode of a fault rotor system according to the Intrinsic mode function (IMF), the installation of the experiment system, the research of principles on the vibration mode of a rotor system under misalignment and rotor-to-stator rub fault, and the development of new time-frequency extraction methods based on improved Hilbert-Huang transform (HHT). All the theoretical approaches listed above were testified by the industrial data obtained from the developed software.The scheme of this study is organized as follows:In Chapter1, rotor’s fault feature extraction methods together with their advantages and disadvantages, were firstly discussed. After the analysis of the significance for the introduction of HHT based feature extraction method, the deficiency of HHT and the necessity for the utilization of modal analysis were presented. The support vector machine (SVM) was here applied to the identification for the typical rotor faults as it has several advantages in the analysis of small fault sample data. As the traditional modal analysis methods have their deficiencies in rotating machinery modal analysis, utilization of operational modal analysis (OMA) was further suggested for fault extraction. Based upon the above analysis, the framework of HHT based fault mode analysis, together with the background, the innovation points and the scheme of this thesis was given. In Chapter2, all the existing fault diagnosis methods were compared from the fault mechanism side. After the definition of fault vibration mode and the introduction of the test-bench, the basic principles of EMD, together with the significance of interpretating Intrinsic mode function’s (IMF’s) physical meanings was addressed. The suitability of interpretating IMF’s physical meaning using OMA technique is also proved. Studying the vibration mode of a typical rotor fault, i.e., rotor-to-stator rub, led to the proposal of the general process for extracting typical rotor fault vibration modes.In Chapter3, the state-of-art of the rotor misalignment fault was reviewed and all the existing fault diagnosis methods were compared. Comparisons impelled the findings of feature extraction problems both under the sub-harmonic and non-sub-harmonic resonance condition. Thus based upon the similarity of frequency composition between misaligned and cracked rotor under the sub-harmonic resonance condition, a full spectrum analysis method was proposed. Also, a newly developed natural frequency extraction method was proposed to make up for the deficiency of non-sub-harmonic feature extraction.In Chapter4, the state-of-art of the rotor-to-stator rub fault was reviewed and all the existing fault diagnosis methods were compared. Based on the analysis, certain improvement had to be made for the original HHT. The improvement is achieved by a newly developed time-frequency analysis tools, named the bandpass filtered HHT (BF-HHT). With this tool, accurate instantaneous frequency can be obtained, which do great help to characterize highly nonstationary signals. Aslo, a new time-frequency extraction procedure, named the spectral kurtosis HHT (SK-HHT), was proposed to automatically extract the most nonstationary frequency components in a signal, in which the nonstationary components usually mean some fault that had happened in the rotor system.In Chapter5, reviewing of the existing pattern recognition methods was firstly made. Considering the advantages of SVM to analyze small sample data, SVM is extremely suitable to be applied to analyze hardly accessible fault data. Study founds out that their exist two major difficulties in fault identification. One is the discrimination between full annular rub and local rub fault. This chapter proposed a effective solution to solve this problem based on EMD-SVD-SVM. The other difficulty is the discrimination of high-feature-frequency characterized faults, i.e., misalignment, crack and initial bow. Based upon the feature extraction of such faults, a new identification procedure is proposed based on Multivariate EMD (MEMD), independent component analysis (ICA) and SVM.In Chapter6, the constitution of condition monitoring software using HHT for supercritical steam turbine generator unit was firstly introduced. Furthermore, the industrial on-line data was obtained and analyzed by this software using the above analysis methods.In Chapter7, the conclusions were made and some advanced topics that need further investigation were presented.